Pain rehabilitation: E/Motion-based automated coaching
Lead Research Organisation:
Imperial College London
Department Name: Computing
Abstract
Abstracts are not currently available in GtR for all funded research. This is normally because the abstract was not required at the time of proposal submission, but may be because it included sensitive information such as personal details.
Organisations
People |
ORCID iD |
Maja Pantic (Principal Investigator) |
Publications
Kaltwang S
(2015)
Latent trees for estimating intensity of Facial Action Units
Liwicki S
(2015)
Online kernel slow feature analysis for temporal video segmentation and tracking.
in IEEE transactions on image processing : a publication of the IEEE Signal Processing Society
Liwicki S
(2014)
Full-Angle Quaternions for Robustly Matching Vectors of 3D Rotations
Marras I
(2014)
Online learning and fusion of orientation appearance models for robust rigid object tracking
in Image and Vision Computing
Martinez B
(2013)
Local evidence aggregation for regression-based facial point detection.
in IEEE transactions on pattern analysis and machine intelligence
Martinez B
(2015)
Facial landmarking for in-the-wild images with local inference based on global appearance
in Image and Vision Computing
Nikitidis S
(2014)
Merging SVMs with Linear Discriminant Analysis: A Combined Model
Description | (a) Pain intensity, as shown in rehabilitation-related scenarios, can be automatically estimated from facial expressions with high Pearson correlation coefficient (CORR >= 0.5). This can be done either by firstly recognising facial actions (i.e. facial action units) underlying the expression of pain, or by estimating the intensity of facial expression of pain directly from the extent of changes in facial features such as the displacement of facial characteristic points. (b) The best results are achieved if accurate facial point trackers are used and facial point locations and displacements are used to represent changes in the observed facial expressions. (c) Discriminative machine learning approaches perform robustly for the target problem (i.e. pain intensity estimation) but cannot handle missing data, which is typical in real-world scenarios as occlusions and self-occlussions often occur. For this problem, it has been shown that a generative approach (i.e. newly-proposed Latent Trees) has a superior performance. |
Exploitation Route | Some of the developed methodologies are publicly available in http://ibug.doc.ic.ac.uk/resources |
Sectors | Digital/Communication/Information Technologies (including Software),Healthcare |
URL | http://www.uclic.ucl.ac.uk/people/n.berthouze/EPain.html |
Description | The consortium collected a large database of multimodal recordings of human behaviour in rehabilitation scenario in which they experienced pain while performing rehabilitation exercises. The database has been properly documented, annotated in terms of pain level as judged by human experts, and released according to ethical clearance guidelines. This database has a very large potential impact as it allows academics and scientists all over the world to study the problem of pain estimation by humans and machines based on various signals including facial expressions captured at a very high frequency and resolution. |
First Year Of Impact | 2015 |
Sector | Digital/Communication/Information Technologies (including Software),Healthcare |
Impact Types | Societal |